Introduction
Underwriters are the backbone of a sound lending process. They make sure that the risk taken (and every form of lending is a risk) is within the appetite of the firm. They also look at the other numerous checks that protect the process and the firm. The underwriting process is subtly and also vastly different depending upon which lending type we look at. For example, a residential mortgage underwriting needs to evaluate any additional borrowing that may be taking place in case of a refinance. A credit card underwriting will not involve that process but would rely more on the credit behavior of the customer. But with all these differences, there are some similarities with an overarching structure to the process. These similarities give us opportunities to assist the rigorous job underwriters have with the power of AI/ML.
More about the automated underwriting process
The first thing to be clear about is that there is no replacement of the underwriter. It is all but known that no amount of credit scoring, file verifications, AML checks are enough to predict the risk effectively. Seasoned underwriters understand the nuances that our systems might miss. They can think of triggering a last-minute re-scoring of the application or asking for just an extra month of payslips etc. We need that ingenuity and in fact, using AI/ML is only going to encourage it.
Using thorough data analysis, it is possible to ink out patterns of borrowing activity. Servicing systems generally save all borrowing data and subsequent payment performance for reporting purposes. It is this data that can be harnessed to generate important insights for a particular case. For example, a certain firm might find that customers who borrow a month from or past Christmas show higher signs of payment difficulty, despite having no issues in their application. Is it because of higher spending around that time that they need them months to adjust with? That is a question that financial experts can answer. But the job of the tech is to show that such a pattern may exist to begin with.
Imagine the power with the underwriter’s disposal, if a similar customer like in our example has a little warning ticker for the underwriter saying, “This customer may face payment difficulties”. That should be enough for the underwriter to go on a more thorough research before signing it off. Or the underwriter may choose to change the product just in case. There are numerous possibilities depending upon how the data pans out.
More applications of automatic underwriting
Automatic underwriting also has application in the AML world. Most AML checks are with historical evidence- by looking at fraud registries and flagging a person previously known to have engaged in suspicious activity. But AI/ML can to a great degree, successfully predict fraudulent behavior. The concept is again the same- using the data which we already painstakingly store. The patterns are what may surprise us and show us insights that we never thought of incorporating in the process.
These are just a few examples of how AI/ML can successfully create powerful interventions in the underwriting world, leading to an automated and more thorough underwriting process.